Win an NVIDIA DGX Spark

Entry status

Entries are open until November 22, 2026 at 11:59 PM CST.

Proposed close

November 22, 2026 at 11:59 PM CST

Proposed grace ends

November 23, 2026 at 11:59 PM CST

Proposed draw

December 1, 2026 at 12:00 PM CST

Eligibility: 18+ and the age of majority where you live. US: all 50 states and DC; Canada excluding Quebec; England, Scotland and Wales; Singapore; Germany; France; Japan. Final eligibility is subject to approved official rules.

Enter the giveaway

The plain version

• One factory-sealed NVIDIA DGX Spark (128 GB, 4 TB), approximate retail value US$6,950. No laptop included.
• For individuals 18+ (and the local age of majority) resident in all 50 US states and DC; Canada excluding Quebec; England, Scotland and Wales; Singapore; Germany; France; Japan. The approved Official Rules control; entries open only when the campaign launches.
• Free to enter. One base entry per eligible person, plus up to 25 bonuses for accepted eligible referrals.
• A referral earns a bonus only after the referred person verifies their email and their entry is accepted. Entering creates your free RuntimeWire account and subscribes you to the newsletter; unsubscribing never affects your entry.
• The winner is chosen at random from verified entries. RuntimeWire staff and their households cannot win.

Full terms: Official Rules. Privacy: Giveaway privacy notice. Questions: Help.

The prize: NVIDIA DGX Spark specifications

Unified memory

128 GB. LPDDR5x memory shared by CPU and GPU.

NVMe storage

4 TB. The configuration included in this giveaway.

Grace Blackwell

GB10. NVIDIA GPU and 20-core Arm CPU.

Memory bandwidth

273 GB/s. NVIDIA's published hardware specification.

Hardware specs: NVIDIA DGX Spark hardware docs. Memory is shared by CPU and GPU, not dedicated VRAM. Product reference photos are not our prize unit; the prize ships to the winner factory sealed and no laptop is included.

Frequently asked questions

Do I need to buy anything or subscribe?

No purchase is needed. Entry is free, and buying RuntimeWire Pro or anything else does not improve your chances. Entering subscribes you to the RuntimeWire newsletter; you can unsubscribe anytime and keep your entry.

Do I need an account?

Entering creates one for you: we verify your email with a one-time code, which signs you in to a free RuntimeWire account. If you are already signed in, your existing account is used.

I clicked the link in the email. Am I entered?

Not yet. Opening a link never enters you. Return to your dashboard and press Confirm my entry.

How do bonus entries work?

Share your personal link from your dashboard. Each unique eligible person who uses it, verifies their email and has an accepted entry earns you one bonus. New bonuses are referral-only, up to 25; the maximum is 26 total entries. Previously awarded bonuses remain valid and count toward the same cap.

Why is my referral pending?

A referral counts only after the person you referred verifies and their entry is accepted.

Who can enter?

Individuals 18 or older and at least the age of majority where they live, legally resident in all 50 US states and DC; Canada excluding Quebec; England, Scotland and Wales; Singapore; Germany; France; Japan. The approved Official Rules control; entries open only when the campaign launches.

My entry was marked ineligible.

You can send an appeal from your dashboard. A person reviews every appeal.

Will the prize be opened or tested?

No. The unit is shipped to the winner factory sealed. We publish no benchmarks or demos of the prize unit.

What people run on one DGX Spark

Not our benchmarks. Published third-party results on a single DGX Spark, not tests of our prize. Different runtimes, quantization and workloads are not a head-to-head comparison or a performance guarantee. Our unit stays sealed.

• Qwen3-30B-A3B (Qwen, 30B total · 3B active): 74.2 decode tokens/s reported, NVFP4 on vLLM. Community-reported decode throughput after a fixed 4K context fill; one DGX Spark. Throughput is not a measure of tool-use correctness. Source: SparkBench, July 10, 2026.
• gpt-oss-120b (OpenAI, 120B): 41.14 decode tokens/s reported, MXFP4 on Ollama 0.12.6. One DGX Spark; firmware 580.95.05. Ten runs, temperature 0, 500-token output cap, caching disabled; book-summary prompt. Source: Ollama, October 23, 2025.
• gpt-oss-20b (OpenAI, 20B): 58.27 decode tokens/s reported, MXFP4 on Ollama 0.12.6. One DGX Spark; firmware 580.95.05. Ten runs, temperature 0, 500-token output cap, caching disabled; book-summary prompt. Source: Ollama, October 23, 2025.
• Qwen3 32B (Qwen, 32B): 9.411 decode tokens/s reported, q4_K_M on Ollama 0.12.6. One DGX Spark; firmware 580.95.05. Ten runs, temperature 0, 500-token output cap, caching disabled; book-summary prompt. Source: Ollama, October 23, 2025.
• DeepSeek-R1 14B (DeepSeek, 14B): 19.99 decode tokens/s reported, q4_K_M on Ollama 0.12.6. One DGX Spark; firmware 580.95.05. Ten runs, temperature 0, 500-token output cap, caching disabled; book-summary prompt. This is the 14B variant, not the full-size model. Source: Ollama, October 23, 2025.
• Gemma 3 12B (Google, 12B): 24.25 decode tokens/s reported, q4_K_M on Ollama 0.12.6. One DGX Spark; firmware 580.95.05. Ten runs, temperature 0, 500-token output cap, caching disabled; book-summary prompt. Source: Ollama, October 23, 2025.